Here's what current and former OpenAI employees are saying about the company's Pentagon deal

Employees of OpenAI are expressing their opinions regarding the company's recent partnership with the Pentagon. Over the weekend, both current and former team members took to X to discuss the implications of this agreement, questioning whether it undermines OpenAI's commitment to safety principles in its dealings with the US Department of Defense. Last week, Sam Altman confirmed that OpenAI has allowed the Defense Department access to its AI models, a move that has drawn comparisons to rival Anthropic, which declined a similar arrangement due to concerns over potential use for domestic surveillance or autonomous weaponry. OpenAI asserted in a blog post that its deal is more robust and includes enhanced safety measures compared to Anthropic’s original contract. In response to growing unease about the implications of the agreement, Altman stated on X that OpenAI is collaborating with the Pentagon to integrate additional safeguards. Boaz Barak, a technical staff member at OpenAI and a professor at Harvard, countered the notion that the company had compromised its safeguards. He argued on X that it is misleading to equate OpenAI's contract with the one rejected by Anthropic, emphasizing that the guarantees against misuse of AI for domestic surveillance or lethal applications are likely stronger in OpenAI's agreement. He added, "The red line of not using AI for domestic mass surveillance should be a shared concern for all of us." Former OpenAI policy research head Miles Brundage shared his apprehensions on X, suggesting that employees should be cautious about the narrative surrounding the deal, implying that OpenAI may have compromised its position while portraying it otherwise. He acknowledged the complexity within OpenAI and the efforts made by many involved in the negotiations. Clive Chan, another OpenAI technical staff member, expressed confidence that the contract prevents the use of models for harmful purposes and urged for transparency regarding the agreement. He recognized the constraints on public disclosures about defense contracts but believed the company should have proactively addressed public concerns. Research scientist Mohammad Bavarian commented that the gap between Anthropic's requirements and the Pentagon’s demands is surmountable, criticizing the Pentagon's labeling of Anthropic as a supply chain risk as unwarranted. Noam Brown, also a researcher at OpenAI, pointed out that the initial language of the agreement raised significant questions about AI's role in lawful surveillance, but noted that recent updates addressed some of these concerns. He stressed the importance of ensuring that democratic processes are upheld in decisions involving AI deployment to intelligence agencies.